Self-similarity and scaling
Self-similarity means that related structure returns when scale changes. The relationship may be an exact copy, a distorted copy or a statistical law rather than a visible motif.
Four kinds of resemblance
Exact self-similarity appears in ideal constructions such as the Sierpiński triangle. Quasi-self-similarity permits distorted copies, as around the Mandelbrot set. Statistical self-similarity preserves distributions rather than individual shapes. Self-affinity scales different directions by different amounts and is important for surfaces and time series.
The four variants answer different questions. In an exact construction, the component copies can be named and their scale calculated. In a self-affine landscape, height and horizontal distance scale differently; uniform resizing would misrepresent its roughness. Statistical self-similarity concerns distributions of sizes, gaps or fluctuations instead. Naming the variant also states what is expected to persist under magnification: a motif, a ratio or a probability law.
A rule, not an optical illusion
A pattern of nested decorative shapes can look fractal without obeying a stable scaling relation. Conversely, noisy data may be self-similar even when the eye sees no repeated icon. The useful question is what measurement transforms predictably with scale and across which interval.
That is why a before-and-after picture is rarely enough. A defensible analysis compares several scales, chooses a measurable quantity and looks for an approximately linear interval on logarithmic axes. A counter-test matters just as much: does the relationship survive changes in crop, resolution or threshold? This robustness distinguishes a scaling law from a coincidental visual effect. Human vision is excellent at finding patterns and equally capable of overestimating repetition.
The natural cutoff
Physical growth, material grain, gravity, diffusion and biological constraints introduce lower and upper cutoffs. A fern leaflet resembles a frond for several levels, not forever. Reporting that range is part of the result, not an inconvenience to be hidden.
The upper cutoff often comes from the total size of the system; the lower one from cells, grains, viscosity or measurement resolution. A wide and practically useful interval can still lie between them. A tree crown need not remain self-similar down to molecules for its branch orders to be meaningfully compared. Good reporting therefore gives not only an exponent, but also the observed scale range and the deviations near its boundaries.
Sources and further reading
This article summarizes the following specialist sources in original wording. Accessed and editorially reviewed 12 August 2026.
- FractalWolfram MathWorld
- Fractal Geometry: Mathematical Foundations and ApplicationsWiley
Continue the idea
What is a fractal?
A clear introduction to fractals, self-similarity, scaling and fractal dimension—with mathematical examples and careful comparisons to nature.
Iteration and recursion
How repeating a rule produces fractal detail, why iteration is not the same as recursion and how limits enter the definition.
Trees, ferns and branching plants
How repeated branching and developmental constraints create approximate self-similarity in plants.